CN-122021914-A - Intelligent agent dialogue generation method and system

CN122021914ACN 122021914 ACN122021914 ACN 122021914ACN-122021914-A

Abstract

The application discloses a method and a system for generating an agent dialogue, and relates to the technical field of agent dialogue generation. The method comprises the steps of obtaining historical dialogue information of a user and an agent dialogue, evaluating and obtaining an acceptance coefficient of the user for inputting dialogue interruption according to the historical dialogue information and recording the acceptance coefficient as a user interruption coefficient, obtaining real-time input content of the user, identifying and obtaining real-time error content according to the real-time input content of the user, obtaining error information of the real-time error content, determining an error influence coefficient according to the error information, obtaining historical error correction information of the user, estimating and obtaining an instant correction coefficient, obtaining a real-time interruption coefficient by combining the error influence coefficient, comprehensively obtaining the agent interruption coefficient by utilizing the user interruption coefficient and the real-time interruption coefficient, determining interruption time according to the agent interruption coefficient, and generating the agent dialogue according to the interruption time. The application improves the efficiency of agent dialogue generation.

Inventors

LIN JIAXIN
Huang Hanye
HUANG GUANGJING
LI RUIQI
HUANG YUETIAN
YANG YONGJIAO
YAN YUPING
ZHU ZEQI
HUANG SHUWEI
LU QING
QIN QIANG

Assignees

广东电网有限责任公司
广东电网有限责任公司数智运营中心

Dates

Publication Date: 20260512
Application Date: 20260203

Claims (10)

1. An agent dialogue generation method, which is characterized by comprising the following steps: Acquiring historical dialogue information of a user and an agent dialogue, evaluating according to the historical dialogue information to obtain an acceptance coefficient of the user for inputting dialogue interruption and recording the acceptance coefficient as a user interruption coefficient; acquiring real-time input content of a user, and identifying real-time error content according to the real-time input content of the user; Obtaining error information of real-time error content, and determining an error influence coefficient according to the error information; acquiring historical error correction information of a user, estimating according to the historical error correction information to obtain an instant correction coefficient, and combining the error influence coefficient to obtain a real-time interrupt coefficient; And comprehensively obtaining an agent breaking coefficient by using the user breaking coefficient and the real-time breaking coefficient, determining breaking time according to the agent breaking coefficient, and generating an agent dialogue according to the breaking time.
2. The method for generating an agent dialogue according to claim 1, wherein the step of obtaining the history dialogue information of the user and the agent dialogue, and evaluating the acceptance coefficient of the user for the input dialogue interruption based on the history dialogue information and recording the acceptance coefficient as the user interruption coefficient is specifically as follows: Acquiring historical dialogue information, counting the number of times of interrupt receiving of a user after the user input is interrupted according to the historical dialogue information, and marking the number of times of interrupt receiving as the interrupt receiving duty ratio; counting the down-frequency of using the agent dialogue after the user input is interrupted and marking as the interrupted down-frequency; acquiring input information after user input is broken and marking the input information as information after breaking, wherein the information after breaking comprises a speech and a speech speed; acquiring input information before the user inputs the interruption, recording the input information as information before the interruption, and comparing the information before the interruption and the information after the interruption to obtain the similarity of the interruption; The basic acceptance coefficient of the user for inputting the conversation interruption is obtained by combining the interruption acceptance proportion, the interruption down-frequency and the interruption similarity; And acquiring reading information of the interrupt output content of the intelligent agent, evaluating according to the reading information to obtain a thinking acceptance coefficient, and combining the basic acceptance coefficient to obtain a user interrupt coefficient.
3. The method for generating a dialogue of an agent according to claim 2, wherein the step of collecting reading information of the interrupt output content of the agent and evaluating the thought acceptance coefficient based on the reading information comprises the steps of: Acquiring complete content corresponding to the interrupt output content of the intelligent agent, and counting the average output percentage of the interrupt output content; collecting reading information of a user for interrupting output content, wherein the reading information comprises average reading duration and average reading quantity; and counting the adoption rate of the user for interrupting the output content, and comprehensively evaluating the average output percentage, the average reading duration and the average reading quantity to obtain the thinking acceptance coefficient.
4. The method for generating an agent dialogue according to claim 1, wherein the step of acquiring error information of real-time error contents and determining an error influence coefficient based on the error information comprises the steps of: judging whether the real-time error content has a real instance or not according to the error information, and if so, acquiring real index information corresponding to the real-time error content; Judging whether the real-time error content is determined to occur or not according to the actual index information, if the real-time error content cannot be determined to occur, estimating the actual occurrence probability of the real-time error content, and estimating to obtain an error occurrence coefficient according to the actual occurrence probability; judging whether the real-time error content is determined to be independent error content according to the error information, and if so, evaluating an error degree coefficient of the real-time error content according to the error information; if the error content is not determined to be independent error content, acquiring associated content of the real-time error content, estimating coverage range values of the error content according to the associated content, and obtaining error influence coefficients by combining error degree coefficients and error occurrence coefficients.
5. The method for generating a dialogue of an agent according to claim 4, wherein the step of estimating the actual occurrence probability of the real-time error content and obtaining the error occurrence coefficient according to the actual occurrence probability is specifically: extracting a reality index reflecting the occurrence of the reality error content according to the reality index information and obtaining a corresponding reflection probability; acquiring real-time reality indexes, marking the real-time reality indexes reflecting the occurrence of real-time error contents as occurrence indexes, and calculating to obtain basic occurrence probability by combining the reflection probability of the occurrence indexes; Collecting voice information of real-time error content input by a user, evaluating according to the voice information to obtain user panic probability, and combining the basic occurrence probability and the user panic probability to obtain actual occurrence probability of the real-time error content; acquiring an actual result corresponding to the real-time error content, and acquiring a development step from the development of the real-time error content to the actual result; and evaluating according to the development steps to obtain an error rescue coefficient, and combining the actual occurrence probability to obtain an error occurrence coefficient.
6. The method for generating an agent dialogue according to claim 5, wherein the step of obtaining the error rescue coefficient based on the step of developing is performed by: Obtaining corresponding real-time steps according to the real-time error contents, and counting the average success rate of saving the real-time error contents in the real-time steps; Counting the average time length from the real-time step development to the actual result, counting the number of the development steps, and obtaining the error rescue probability by combining the average success rate evaluation; acquiring average execution ratio and average execution accuracy of the user to the output content of the intelligent agent, and evaluating to obtain a user rescue coefficient; and combining the error rescue probability and the user rescue coefficient to comprehensively obtain the error rescue coefficient.
7. The method for generating an agent session according to claim 4, wherein the step of evaluating the error degree coefficient of the real-time error content based on the error information comprises: judging whether the real-time error content appears in the history dialogue, if so, counting the average influence degree of the real-time error content on the result and recording the average influence degree as the result influence degree; counting other dialogue duty ratios of the real-time error content and recording the other dialogue duty ratios as error migration duty ratios; counting the number of times that the user finds the same errors as the real-time error content in the historical dialogue, and recording the number as the found duty ratio; combining the result influence degree, the error migration duty ratio and the discovery duty ratio to obtain an error degree coefficient; if not, the result difference degree of the real-time error content and the corresponding correct content is obtained as an error degree coefficient.
8. The method for generating a dialogue of an agent according to claim 4, wherein the step of collecting the associated content of the real-time error content and estimating the coverage value of the error content according to the associated content comprises the steps of: Acquiring user input content, and searching to acquire core content input by a user according to the user input content; Expanding according to the core content to obtain user discussion content, and searching the occurrence ratio of real-time error content in the user discussion content; collecting associated content of real-time error content, counting content duty ratio of the associated content based on the real-time error content and recording the content duty ratio as base duty ratio; counting the content quantity of the associated content, counting the frequency of simultaneous occurrence of the real-time error content and the associated content, and marking the frequency as the associated frequency; and calculating the coverage range value of the error content by combining the occurrence ratio, the basic ratio, the content quantity and the associated frequency.
9. The method for generating an agent dialogue according to claim 1, wherein the step of obtaining the historical error correction information of the user and estimating the instant correction coefficient according to the historical error correction information comprises the steps of: obtaining the average correction rate and the average correction duration of the same errors of the real-time error content in the historical dialogue by the user, and evaluating to obtain the user error correction coefficient; Comparing the average inner capacity difference of the input contents before and after the user is interrupted by the agent and recording the average inner capacity difference as a capacity difference; comparing the smoothness difference value of the input content before and after the user is interrupted by the intelligent agent, counting the average duration of the input content again after the user is interrupted by the intelligent agent, and recording the average duration as the arrangement duration; And obtaining a broken thinking coefficient by using the capacity difference value, the smoothness difference value and the arrangement time length, and obtaining an instant correction coefficient by combining the user error correction coefficient.
10. An agent session generation system, characterized by applying an agent session generation method according to any one of claims 1-9, comprising: The user interrupt module acquires the historical dialogue information of the dialogue between the user and the intelligent agent, evaluates and obtains the acceptance coefficient of the user for inputting dialogue interrupt according to the historical dialogue information and marks the acceptance coefficient as the user interrupt coefficient; the real-time error module is used for acquiring real-time input content of a user and identifying the real-time error content according to the real-time input content of the user; the error influence module acquires error information of the real-time error content and determines an error influence coefficient according to the error information; The real-time breaking module is used for acquiring the historical error correction information of the user, obtaining an instant correction coefficient according to the estimated historical error correction information, and obtaining a real-time breaking coefficient by combining the error influence coefficient; And the dialogue generation module is used for comprehensively obtaining the agent breaking coefficient by utilizing the user breaking coefficient and the real-time breaking coefficient, determining breaking time according to the agent breaking coefficient and generating the agent dialogue according to the breaking time.

Description

Intelligent agent dialogue generation method and system Technical Field The application relates to the technical field of agent dialogue generation, in particular to an agent dialogue generation method and system. Background In the development of the current intelligent agent dialogue generation technology, the existing method mainly focuses on improving the relativity, individuation and diversity of dialogue contents, but generally ignores the intelligent adaptation problem of a breaking mechanism in the dialogue interaction process. Conventional dialog systems typically employ sequential interaction modes, rely on predefined dialog flows, semantic understanding models, and context tracking algorithms, and generate corresponding replies after complete parsing of the content of the user-entered sentence. However, in real-time interaction scenarios (such as online customer service, voice assistant and virtual conference), due to the instantaneity and the dynamics of the dialogue, if the system only passively waits for the user to finish inputting and then responds, the problems of tuina of dialogue rhythm, high response delay, low interaction efficiency and the like are caused, so that not only is the error content caused to occupy a large amount of dialogue time, but also the adaptability and the initiative of the intelligent agent in the dynamic dialogue are limited. Disclosure of Invention The invention aims to provide an agent dialogue generating method and system for solving the problems in the background technology. In a first aspect, the present application provides a method for generating an agent dialogue, which adopts the following technical scheme: Acquiring historical dialogue information of a user and an agent dialogue, evaluating according to the historical dialogue information to obtain an acceptance coefficient of the user for inputting dialogue interruption and recording the acceptance coefficient as a user interruption coefficient; acquiring real-time input content of a user, and identifying real-time error content according to the real-time input content of the user; Obtaining error information of real-time error content, and determining an error influence coefficient according to the error information; acquiring historical error correction information of a user, estimating according to the historical error correction information to obtain an instant correction coefficient, and combining the error influence coefficient to obtain a real-time interrupt coefficient; And comprehensively obtaining an agent breaking coefficient by using the user breaking coefficient and the real-time breaking coefficient, determining breaking time according to the agent breaking coefficient, and generating an agent dialogue according to the breaking time. Preferably, the step of obtaining the historical dialogue information of the dialogue between the user and the agent, and evaluating the acceptance coefficient of the user for the input dialogue interruption according to the historical dialogue information and marking the acceptance coefficient as the user interruption coefficient is specifically as follows: Acquiring historical dialogue information, counting the number of times of interrupt receiving of a user after the user input is interrupted according to the historical dialogue information, and marking the number of times of interrupt receiving as the interrupt receiving duty ratio; counting the down-frequency of using the agent dialogue after the user input is interrupted and marking as the interrupted down-frequency; acquiring input information after user input is broken and marking the input information as information after breaking, wherein the information after breaking comprises a speech and a speech speed; acquiring input information before the user inputs the interruption, recording the input information as information before the interruption, and comparing the information before the interruption and the information after the interruption to obtain the similarity of the interruption; The basic acceptance coefficient of the user for inputting the conversation interruption is obtained by combining the interruption acceptance proportion, the interruption down-frequency and the interruption similarity; And acquiring reading information of the interrupt output content of the intelligent agent, evaluating according to the reading information to obtain a thinking acceptance coefficient, and combining the basic acceptance coefficient to obtain a user interrupt coefficient. Preferably, the step of acquiring reading information of the interrupt output content of the intelligent agent and evaluating and obtaining the thinking acceptance coefficient according to the reading information comprises the following specific steps: Acquiring complete content corresponding to the interrupt output content of the intelligent agent, and counting the average output percentage of the interrupt output content; collecting re